Papers › HyperSeg: Patch-wise Hypernetwork for Real-time Semantic Segmentation

HyperSeg: Patch-wise Hypernetwork for Real-time Semantic Segmentation

21 Dec 2020CVPR 2021 1arXiv:2012.11582archive 2025-07-28

Yuval Nirkin, Lior Wolf, Tal Hassner

We present a novel, real-time, semantic segmentation network in which the encoder both encodes and generates the parameters (weights) of the decoder. Furthermore, to allow maximal adaptivity, the weights at each decoder block vary spatially. For this purpose, we design a new type of hypernetwork, composed of a nested U-Net for drawing higher level context features, a multi-headed weight generating module which generates the weights of each block in the decoder immediately before they are consumed, for efficient memory utilization, and a primary network that is composed of novel dynamic patch-wise convolutions. Despite the usage of less-conventional blocks, our architecture obtains real-time performance. In terms of the runtime vs. accuracy trade-off, we surpass state of the art (SotA) results on popular semantic segmentation benchmarks: PASCAL VOC 2012 (val. set) and real-time semantic segmentation on Cityscapes, and CamVid. The code is available: https://nirkin.com/hyperseg.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

YuvalNirkin/hyperseg officialmentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

DecoderDichotomous Image SegmentationReal-Time Semantic SegmentationSegmentationSemantic Segmentation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Dichotomous Image Segmentation DIS-TE1 HySM E-measure 0.803 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE1 HySM HCE 205 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE1 HySM MAE 0.082 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE1 HySM S-Measure 0.761 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE1 HySM max F-Measure 0.695 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE1 HySM weighted F-measure 0.597 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM E-measure 0.832 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM HCE 451 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM MAE 0.085 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM S-Measure 0.794 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM max F-Measure 0.759 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE2 HySM weighted F-measure 0.667 #7 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM E-measure 0.857 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM HCE 887 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM MAE 0.079 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM S-Measure 0.811 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM max F-Measure 0.792 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE3 HySM weighted F-measure 0.701 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM E-measure 0.842 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM HCE 3331 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM MAE 0.091 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM S-Measure 0.802 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM max F-Measure 0.782 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-TE4 HySM weighted F-measure 0.693 #8 of 22 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM E-measure 0.814 #10 of 24 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM HCE 1324 #10 of 24 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM MAE 0.096 #10 of 24 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM S-Measure 0.773 #10 of 24 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM max F-Measure 0.734 #10 of 24 Archive leaderboard report
Dichotomous Image Segmentation DIS-VD HySM weighted F-measure 0.640 #10 of 24 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-L Frame (fps) 16.6 #7 of 29 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-L Time (ms) 60.2 #7 of 29 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-L mIoU 79.1 #7 of 29 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-S Frame (fps) 38.0 #9 of 29 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-S Time (ms) 26.3 #9 of 29 Archive leaderboard report
Real-Time Semantic Segmentation CamVid HyperSeg-S mIoU 78.4 #9 of 29 Archive leaderboard report
Real-Time Semantic Segmentation Cityscapes test HyperSeg-M Frame (fps) 36.9 #11 of 39 Archive leaderboard report
Real-Time Semantic Segmentation Cityscapes test HyperSeg-M Time (ms) 27.1 #11 of 39 Archive leaderboard report
Real-Time Semantic Segmentation Cityscapes test HyperSeg-M mIoU 75.8% #11 of 39 Archive leaderboard report
Semantic Segmentation PASCAL VOC 2012 val HyperSeg-L mIoU 80.61% #9 of 29 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Concatenated Skip ConnectionConvolutionHyperNetworkMax PoolingReLUU-Net

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections